config.yaml

base: &BASE

    dataname: 'raw' # base name of training/validation data
    img_dim: 128 # image dimension
    noise_vect_len: 64 # length of generator noise vector
    nlayers: 4 # number of conv+batchnorm layers
    convkern: 5 # convolution kernel size
    convstride: 2 # convolution kernel stride
    nconvfilters: [64, 128, 256, 512] # number of convolutional filters per layer in discriminator
    ndeconvfilters: [256, 128, 64, 1] # number of deconvolutional filters per layer in generator
    label_flip: 0.01 # probability of flipping real.fake labels
    batchsize: 64
    print_batch: 20 # print losses every n batches
    checkpt_batch: 150 # save tensorboard checkpoint every n batches
    cscale: 'lin' # colorscale for image plots
    datascale: 4 # scaling parameter for normalizing data
    learn_rate: [0.0002, 0.0002] # learning rates for [G, D]
    DG_update_ratio: 2 # number of D updates per G update
    Nepochs: 200 
    datafmt: 'channels_first' # 'channels_first' for NCHW, 'channels_last' for NHWC
    multichannel: True # enable/disable multi-channel rescaling